Increased nuclear factor I-mediated chromatin access drives transition to androgen receptor splice variant dependence in prostate cancer

SUMMARY

Androgen receptor (AR) splice variants, of which ARv7 is the most common, are increased in castration-resistant prostate cancer, but the extent to which they drive AR activity is unclear. We generated a subline of VCaP cells (VCaP16) that is resistant to the AR inhibitor enzalutamide (ENZ). AR activity in VCaP16 is driven by ARv7, independently of full-length AR (ARfl), and its cistrome and transcriptome mirror those of ARfl in VCaP cells. ARv7 expression increases rapidly in response to ENZ, but there is a delay in gaining chromatin binding and transcriptional activity, which is associated with increased chromatin accessibility. AR and nuclear factor I (NFI) motifs are most enriched at more accessible sites, and NFIB/X knockdown greatly diminishes ARv7 function. These findings indicate that ARv7 can drive the AR program but that its activity is dependent on adaptations that increase chromatin accessibility to enhance its intrinsically weak chromatin binding.

In brief

Poluben et al. show that prostate cancer cell resistance to an androgen receptor (AR) antagonist targeting the full-length AR (ARfl) is driven by an AR spice variant (ARv7) that functions independently of ARfl but requires adaptations associated with increased chromatin accessibility and is dependent on nuclear factor I transcription factors.

Graphical abstract

INTRODUCTION

Prostate cancer (PC) that recurs after androgen deprivation therapy (castration-resistant prostate cancer [CRPC]) is in most cases still dependent on androgen receptor (AR) activity driven by residual androgens.^1–3 AR activity in CRPC can be further suppressed by abiraterone or AR antagonists such as enzalutamide (ENZ).^4–6 Unfortunately, patients treated with these androgen-signaling inhibitor (ASI) drugs invariably progress. A subset become AR independent and may undergo neuroendocrine differentiation,^7,8 but most continue to express high levels of AR,^9–11 with persistent AR transcriptional activity. A subset has AR ligand-binding domain (LBD) mutations that allow alternative steroids^12–14 or antagonists to act as agonists.^15,16 Activation of several signaling pathways also may directly or indirectly enhance AR activity.¹⁷ Finally, increasing evidence indicates that AR splice variants that are constitutively active due to lack of the LBD, of which AR variant 7 (ARv7) is the most common, contributes to persistent AR activity in tumors that progress on ASI therapy.^18–23

ARv7 is generated by splicing to a cryptic exon downstream of exon 3 (encoding the DNA-binding domain), resulting in an AR that is truncated after the DNA-binding domain. ARv7 is low in untreated primary PC but is expressed in most cases of CRPC and is further increased in ASI-resistant tumors.^21,22 ARv7 can drive AR activity in cells where ARfl activity is suppressed.^24–27 Moreover, ARv7 in circulating tumor cells is associated with resistance to ASI drugs, suggesting it is a mediator of resistance.^19,23,28 However, most tumors expressing ARv7 also express high levels of ARfl, so the extent to which ARv7 is an ARfl-independent driver of AR activity versus a biomarker of tumors that are more aggressive and resistant due to other mechanisms, including high levels of ARfl, remains unclear. To address the role of ARv7 we have characterized ENZ-resistant VCaP PC cells that express both high levels of ARfl and increased levels of ARv7. We find that ARv7 drives ARfl-independent activity but is dependent on adaptations that enhance chromatin accessibility.

RESULTS

ENZ resistance in VCaP cells is associated with AR reactivation and expression of ARv7

VCaP cells were cultured in medium with increasing concentrations of ENZ (up to 16 μM) over ~2 months to select for ENZ resistance (VCaP16 cells). These were then compared with parental VCaP cells (grown in medium with 10% fetal bovine serum [FBS]) and with VCaP cells exposed short term (4 days) to 16 μM ENZ (added to medium with 10% FBS). Addition of ENZ initially suppressed VCaP proliferation, but proliferation of VCaP16 (maintained on 16 μM ENZ) was comparable to that of the parental VCaP cells (Figure 1A). As expected, short-term ENZ treatment of VCaP cells (VCaP-E) suppressed AR-regulated genes, but their expression was restored in the VCaP16 (Figure S1A).

Figure 1. VCaP16 cells exhibit increased ARv7 expression and resistance to AR blockade.

(A) Proliferation for VCaP, VCaP with 16 μM ENZ (VCaP-E), and VCaP16 (maintained in 16 μM ENZ) at 7 days post treatment (*p < 0.05, **p < 0.01, ANOVA).

(B) Hallmark androgen response comparing VCaP cells with ENZ for 4 days (VCaP-E) or VCaP16 versus parental VCaP. Significantly differentially expressed genes (adjusted p value <0.05) were input here and below. FDR, false discovery rate; NES, normalized enrichment score.

(C) Immunoblot of VCaP16 and parental VCaP.

(D) ARfl and ARv7 assessed by RT-qPCR in VCaP (black) and VCaP16 (red) cells (*p < 0.05, **p < 0.01, unpaired t test, two-tailed).

(E) (Upper) KLK3 by RT-qPCR and (lower) ARfl and ARv7 in VCaP16 cells after siRNA knockdown with non-targeting control (siNTC), ARv7 (siV7), both ARv7 and ARfl (siEx1), or just ARfl (siEx7) siRNA. Bars represent SEM. (*p < 0.05, **p < 0.01, ns = non-significant, Welch’s t test).

(F) Hallmark androgen response in VCaP16 with siRNA knockdown of ARv7 (siV7) or ARfl (siEx7).

(G) VCaP16 cells treated with ARCC-32 (500 nM) or DMSO for 36 h.

(H) Hallmark androgen response in VCaP16 with ARCC-32 (500 nM) or DMSO for 36 h. Input was all differentially expressed genes.

(I) Live-cell DNA fluorescence in VCaP16 treated with non-target control siRNA (siNTC) or ARv7 siRNA (siV7) ± 500 nM ARCC-32.

(J) Differential expression of ARv7 gene set from Sharp et al.^21,22 comparing VCaP16 versus VCaP cells (left) or VCaP16 treated with ARv7 siRNA versus non-target control siRNA (siNTC) (right). Dotted line represents a value of 2 (p = 0.01).

RNA sequencing (RNA-seq) confirmed that VCaP exposed to ENZ for 4 days had marked decreases in hallmark gene sets related to androgen response (Figure 1B, left) and proliferation (Figures S1B and S1C). The MTORC1 gene set was also markedly decreased,²⁹ while no hallmark gene sets were increased. Gene sets related to proliferation and MTORC1 were also decreased in VCaP16 relative to parental VCaP, but to a lesser degree (Figures S1B and S1D). Moreover, VCaP16 exhibited no significant changes in the androgen response gene set compared to VCaP, indicating that AR activity was substantially restored (Figure 1B, right). This restoration was not associated with increased ARfl protein or mRNA (Figures 1C and 1D), and DNA sequencing did not show AR exonic mutations (Figure S1E). In contrast, ARv7 protein and mRNA were increased (Figures 1C and 1D).

AR activity in VCaP16 cells is driven by ARv7

Notably, depletion of ARfl with small interfering RNA (siRNA) against exon 7 (siEx7) did not decrease AR-regulated KLK3/prostate-specific antigen (PSA) (Figure 1E). In contrast, ARv7 (siV7) depletion markedly decreased KLK3, which was also decreased by exon 1 siRNA targeting both ARfl and ARv7 (siEx1). RNA-seq confirmed that ARv7 siRNA, but not ARfl siEx7, markedly decreased hallmark androgen response genes (Figures 1F and S1B). Principal component analysis similarly showed that ARfl knockdown had minimal effects, while the ARv7 and Ex1 siRNA samples were separated from the control siRNA and clustered together (Figures S2A–S2C).

We next treated VCaP16 with an AR degrader (ARCC-32, ARD) targeting the LBD to selectively decrease ARfl protein (Figure 1G). We used a relatively high concentration of ARCC-32 (500 nM) to compete with the ENZ. The AR degrader decreased expression of PSA, possibly reflecting more acute or potent suppression of ARfl than with siEx7. However, RNA-seq showed that overall AR activity was not decreased by the AR degrader, with a trend toward increased activity (Figure 1H). Principal component analysis similarly showed that the AR degrader had minimal overall impact (Figures S2A–S2C). Consistent with these results, we found that VCaP16 proliferation was more suppressed by ARv7 siRNA than the ARfl degrader (Figure 1I). Finally, to assess the extent to which ARv7 activity in VCaP16 is reflective of clinical samples, we examined a panel of genes whose expression was correlated with ARv7 protein expression in clinical CRPC.²¹ Notably, these genes were increased in VCaP16 and decreased by siRNA targeting ARv7 (Figure 1J).

ARv7 has limited chromatin-binding capacity acutely following ENZ treatment

ARv7 was increased in VCaP within 2 days of ENZ treatment, and levels peaked by 7–14 days (Figure 2A).³⁰ However, restoration of AR activity took several more weeks. This appeared to reflect adaptation rather than clonal selection, as whole-exome sequencing did not show differences in copy number or driving mutations (Figure S3). This initial lack of transcriptional activity was not due to decreased ARv7 accumulation in the nucleus, as ARv7 was nuclear in both the short-term ENZ-treated cells (VCaP-E) and VCaP16 (Figure S4A). ARfl was also primarily nuclear in both (Figure S4B). Confocal microscopy further showed that ARv7 was localized primarily in discrete puncta in VCaP-E and VCaP16 (Figure 2B). There was substantial colocalization of CREB-binding protein (CBP) in these structures, consistent with them being transcriptional condensates. Although the number of puncta per cell was greater in VCaP16 (Figure 2C), a similar fraction was colocalized with CBP in both cells (Figure 2D).

Figure 2. ARv7 induced acutely by ENZ is nuclear but lacks chromatin-binding capacity.

(A) VCaP with 16 μM ENZ from 2 to 28 days and VCaP16 cells. AR-NT, AR N terminus antibody (long/short exposures).

(B) Confocal immunofluorescence of ARv7 and CBP/p300 in DMSO or 16 μM ENZ (4 days) treated VCaP versus VCaP16 (maintained in 16 μM ENZ).

(C and D) Quantification of z-stack three-dimensional images showing number of ARv7 puncta per nuclei in four fields (C) and fraction of ARv7 puncta containing CBP/p300 (D).

(E) Chromatin fraction from VCaP and VCaP16 following formalin crosslinking for ChIP.

(F) ARv7-binding sites based on ChIP-seq in samples from (E) and Venn diagram with overlap between ARv7 sites in VCaP16 (ENZ) in biological replicates.

(G) ARfl-binding sites by AR C-terminal antibody ChIP-seq on samples in (E).

We next assessed ARv7 chromatin binding following 3 days in ENZ (VCaP-E) versus VCaP16. Notably, ARv7 protein levels were comparable after formaldehyde crosslinking in the chromatin fraction of VCaP-E and VCaP16 (Figure 2E). In contrast, chromatin immunoprecipitation sequencing (ChIP-seq) showed that ARv7 binding was substantially increased in VCaP16 (2,150 common sites in replicate samples) (Figure 2F). ARfl binding was also greater in VCaP16, although the fold increase was less than for ARv7 (Figure 2G). Notably, the number of identified ARfl sites in VCaP16 (52,536 sites) was much greater than the number of ARv7 sites, which may reflect more binding and/or relative efficiency of the ChIP antibodies (see below). Finally, acute dihydrotestosterone (DHT) treatment (4 h) caused an increase in ARfl binding (as expected) but also increased ARv7 binding in VCaP16, which may reflect mechanisms including ARfl/ARv7 heterodimer formation (see below). Together, these data show that adaptation to ENZ is associated with an increase in site-specific ARv7 chromatin binding.

ARv7 binds chromatin independently of ARfl

AR activity in VCaP16 after ARfl depletion indicated that ARv7 can transactivate independently of ARfl. Consistent with this conclusion, ARfl degradation did not impair ARv7 chromatin binding in VCaP16 (Figure 3A). ChIP-seq showed a small decrease in ARv7 peak intensities (Figure 3B), but the number of sites was not decreased, with a trend toward increased ARv7-binding sites (Figure 3C). Moreover, the majority of the ARv7 sites in the control VCaP16 cells were also in the ARCC-32-treated cells (Figure 3D).

Figure 3. ARv7 chromatin binding in VCaP16 cells is independent of ARfl.

(A) VCaP16 in basal medium (16 μM ENZ) was treated for 24 or 48 h with vehicle (NC) or ARCC-32 (500 nM). Cells were then separated into cytoplasmic, soluble nuclear, or chromatin fractions.

(B) ARv7 binding in VCaP16 in basal medium (16 μM ENZ) with addition of DMSO or ARCC-32 (500 nM) for 36 h. ChIP-seq signal is centered at shared ARv7 sites.

(C) ARv7-binding sites in VCaP16 in basal medium (with 16 μM ENZ) treated with DMSO or ARCC-32 (500 nM) for 36 h.

(D) Intersection of merged ARv7 sites in VCaP16 cells in DMSO or ARCC-32 (500 nM).

(E) VCaP16 were double-crosslinked with DSP and formaldehyde. Chromatin was then isolated and solubilized with benzonase, followed by immunoprecipitation.

(F) ARv7 and ARfl sites in VCaP16 cells following 4 h of treatment with DHT (10 nM).

(G) ARv7 and ARfl binding in VCaP16 following 4-h treatment with DHT (10 nM). ChIP-seq signal is centered at ARv7 sites in DHT-treated VCaP16.

(H) Whole-cell lysate (WCL) and chromatin fraction from VCaP16 in basal medium (16-μM ENZ) or treated for 4 h with DHT (10 nM).

(I) Enrichment of ARv7 binding in LNCaP95 cells in basal medium treated for 2 h with DMSO, DHT (10 nM), or dexamethasone (DEX, 100 nM). Signal centered at ARv7 sites identified in VCaP16 cells.

To further address ARfl dependence, VCaP16 cells were initially treated with ARCC-32, and protein complexes were then stabilized by double fixation with a cleavable crosslinker dithiobis(succinimidyl propionate) (DSP) and formaldehyde, followed by chromatin isolation and DNA digestion with benzonase. ARCC-32 again did not diminish ARv7 associated with chromatin (Figure 3E). ARv7 immunoprecipitation showed trace levels of ARfl that were lost in the ARCC-32-treated cells, which could reflect heterodimers or adjacent homodimers. Notably, ARCC-32 did not decrease HOXB13 or BRG1 coprecipitated with ARv7, indicating that these interactions were not ARfl dependent. Finally, HOXB13 immunoprecipitation demonstrated that ARv7-HOXB13 interaction was ARfl independent. These findings are consistent with our previous results in LNCaP95 and CWR22Rv1, where we similarly found ARfl-independent ARv7 binding to chromatin.³¹

While these data support ARfl-independent binding, studies with ectopically expressed ARv7 clearly show it can both act as a homodimer and heterodimerize with ARfl.^32–35 Similarly to results in LNCaP95 cells,²⁷ ChIP-seq showed that DHT increased ARv7 binding (see Figure 2F), and the gained ARv7-binding sites overlapped ARfl sites (Figure 3F). Moreover, a heatmap of the ChIP-seq data indicates that DHT broadly and proportionately enhances ARv7 binding to sites that are bound more weakly prior to DHT, consistent with ARfl/ARv7 heterodimerization (Figure 3G). However, in contrast to the ChIP-seq results, examination of the chromatin fraction showed that DHT caused a marked increase in ARfl binding but no increase in ARv7 binding, suggesting that alternative mechanisms may also contribute to the increased ARv7 binding seen by ChIP-seq (Figure 3H).

Another mechanism by which ARfl may enhance ARv7 binding is through assisted chromatin loading, whereby binding of an ARfl homodimer makes the site more available for subsequent binding by ARv7. Indeed, a previous study found that glucocorticoid receptor (GR) binding at a glucocorticoid-responsive element enhances binding of a variant GR to the same site.³⁶ To assess this, we took advantage of the overlap between GR and ARv7-binding sites and asked whether GR activation would enhance ARv7 binding at these sites. LNCaP95 cells were treated with vehicle, DHT, or dexamethasone (DEX; a GR-specific agonist), followed by ARv7 ChIP-seq. Consistent with assisted loading, DEX enhanced ARv7 binding at the AR/GR-regulated FKBP5 gene (Figure S4A). Similarly, ARv7 binding was increased by DEX in the AR/GR (and mineralocorticoid receptor) regulated SCNN1A gene. We then examined previous GR ChIP-seq data to identify GR/ARv7 shared sites and found that DEX caused a broad increase in ARv7 binding to these sites (Figures S5B and S5C). This was also observed when we examined all ARv7 sites (Figure 3I). Together, these findings indicate that ARv7 functions independently of ARfl in VCaP16 but in the presence of androgen can function cooperatively with ARfl through heterodimerization, assisted loading, or possibly other mechanisms. The results also suggest that GR, in addition to directly activating a subset of AR-target genes,³⁷ may indirectly enhance ARfl and/or ARv7 activity by assisted loading. Notably, GR mRNA was not increased in VCaP16 cells, while there was a modest increase in mineralocorticoid receptor expression (Figure S5D).

ARv7 cistrome is enriched for higher-affinity ARfl-binding sites

Previous studies have been inconsistent as to whether ARv7 has a distinct cistrome and/or transcriptome.^26,38–42 In LNCaP95 cells, essentially all ARv7-binding sites were also ARfl sites.²⁷ Similarly, ARv7 ChIP-seq in VCaP16 cells revealed ~2,150 peaks, virtually all of which (97%) overlapped with ARfl peaks (Figures 4A and 4B). Notably, while the number of called ARv7 peaks is low relative to ARfl peaks, inspection of heatmaps centered on total ARfl sites showed a correlation between ARfl and ARv7 binding across most or all ARfl peaks (Figure 4C). It is not clear whether the lower number of called ARv7 peaks reflects weaker ARv7 binding to chromatin or lower efficiency of the ARv7 antibody used for ChIP (or both). In support of the former hypothesis, profile plots show that ARfl binding is markedly greater at called ARv7 sites, suggesting that these are being picked up by the ARv7 antibody as they reflect higher-affinity ARfl/ARv7 sites (Figure 4D). Notably, ARfl binding is also greater in VCaP16 relative to VCaP-E cells, suggesting that adaptations in VCaP16 that increased ARv7 binding may also increase ENZ-liganded ARfl binding (Figure 4E).

Figure 4. ARv7- and ARfl-binding sites overlap with proportionate peak intensities.

(A) Intersection of ARv7 and ARfl sites in VCaP16 in basal medium (16 μM ENZ).

(B) ARfl and ARv7 binding at called ARv7 sites (upper) and at ARfl-unique sites.

(C) ARv7- and ARfl-binding intensities across all ARfl sites in VCaP16, and showing ARv7 in VCaP after short-term ENZ (4 days) (VCaP-E).

(D) Peak intensities for ARfl and ARv7 in VCaP16 at ARv7 versus ARfl-unique sites. ChIP-seq signal centered at ARv7/ARfl shared (ARv7) and ARfl-unique (ARfl) sites identified in VCaP16.

(E) ARfl-binding intensities in (1) VCaP16 in basal medium (16 μM ENZ), (2) VCaP16 with DHT (10 nM) for 4 h, (3) VCaP with ENZ for 72 h followed by removal of ENZ and addition of DHT (10 nM) for 4 h, and (4) VCaP with ENZ for 72 h. ChIP-seq signal is centered at total ARfl sites identified in DHT-stimulated VCaP16.

(F and G) Pearson’s correlation of normalized read counts: ARv7 versus ARfl called at shared ARfl/ARv7 and ARfl-unique sites in (F) VCaP16 and (G) VCaP16 treated with DHT.

(H) AR signal enrichment with AR N-terminal (GSM2842708), AR C-terminal (GSM2842700), and ARv7 (GSM2842704) antibodies in LNCaP95 cells, centered at shared ARfl/ARv7 (ARv7) sites and ARfl-unique sites identified in VCaP16.

(I) Pearson’s correlation of normalized read counts: ARv7 (GSM2842704) versus ARfl (GSM2842700) in LNCaP95 at shared ARfl/ARv7 (ARv7) sites and ARfl-unique sites identified in VCaP16.

(J) Top enriched motifs (±100 kb from peak center) at ARfl-unique sites and at shared ARfl/ARv7 sites in VCaP16.

To further assess the relationship between ARv7 versus ARfl binding in VCaP16, we plotted normalized read counts for ARv7 versus ARfl at ARfl-unique sites (sites where ARv7 was not called by the HOMER algorithm) and at ARv7 sites (which are all shared with ARfl) (Figure 4F). Significantly, ARv7 and ARfl binding was highly correlated at ARv7 sites and at ARfl-unique sites, supporting the conclusion that ARv7 is also binding proportionately to ARfl at these latter ARfl-unique sites. We also compared ARv7 versus ARfl binding in VCaP16 cells after DHT treatment and saw a similar strong correlation at ARv7 and ARfl-unique sites (Figure 4G). Interestingly, the slope of the line (ratio of ARfl to ARv7 binding) was greater in the DHT-treated cells, consistent with DHT driving stronger ARfl binding.

We next re-examined previous ARfl and ARv7 ChIP-seq data in LNCaP95 cells.²⁷ Peak intensities with an AR N-terminal antibody (Ab), AR C-terminal Ab, and ARv7 Ab were all markedly greater at VCaP16-identified ARv7 sites versus ARfl-unique sites (Figure 4H). There was also a similar correlation between ARfl- and ARv7-binding intensities at ARfl-unique and ARv7 sites identified in the LNCaP95 cells (Figure 4I). Further AR ChIP-seq data in CWR22RV1, LNCaP95, and LNCaP also showed increased ARv7 binding at the VCaP16 ARv7 sites and indicated that these correspond to high-affinity ARfl-binding sites (Figures S6A–S6C). Moreover, an analysis of normal prostate, primary PC clinical samples, and patient-derived xenograft (PDX) CRPC models also showed increased ARfl binding at these ARv7 sites (Figure S6D).^43,44 Finally, most VCaP16 ARv7-binding sites (89.3%) overlapped ARfl sites that were shared by at least two tumors in an AR ChiP-seq analysis of 88 primary PCs⁴⁵ (Figure S6E).

We then carried out a motif analysis, which that showed the ARv7 sites (relative to ARfl sites) were greatly enriched for the AR/GR/PR motif, consistent with them being high-affinity binding sites (Figures 4J and S7). Together, these results show that the ARv7 and ARfl cistromes in VCaP16 are largely overlapping and also overlap with the cistrome of the DHT-liganded AR. Moreover, they show that ARv7-binding intensity correlates with affinity for the DHT-liganded ARfl, and that called ARv7 sites in VCaP16 are high-affinity ARfl sites across multiple models and clinical samples.

ARv7 transcriptome overlaps with that of ARfl

We next compared the initial effects of ENZ in VCaP cells with siRNA-mediated depletion of ARv7 in VCaP16 cells. This showed a strong correlation between genes regulated by ARfl in VCaP cells and genes regulated by ARv7 in VCaP16 cells, indicating that the ARfl transcriptome in VCaP cells was largely restored by ARv7 in VCaP16 cells (Figure 5A). A previous study in LNCaP95 cells found a marked difference in the ARfl versus ARv7 transcriptomes, with ARv7 acting to suppress expression of a subset of growth-promoting genes.²⁷ Notably, this previous study compared effects of depleting ARfl versus ARv7 in the same cell line, whereas the analysis here compared blocking ARfl in VCaP with ARv7 depletion in VCaP16. Therefore, we next compared the effects of depleting ARfl or ARv7 in VCaP16 and found a much weaker correlation between genes altered by depletion of ARfl (siEx7) versus depletion of ARv7 (siV7), with ARfl depletion increasing expression of many genes that are decreased by ARv7 depletion (Figure 5B). The correlation was even weaker (and became inverse) when the AR degrader ARCC-32 was used to deplete ARfl (Figure 5C). We also compared effects of exon 1 siRNA (targeting ARfl and ARv7) with siRNA targeting ARfl (exon 7 siRNA) (Figure 5D) or ARv7 (Figure 5E). Notably, the strongest correlation was with ARv7 siRNA, further showing that ARv7 is the predominant driver of AR activity. We next looked specifically at genes associated with ARv7-binding sites. These were greatly decreased by siRNA targeting both ARfl and ARv7 (exon 1 siRNA) and by siRNA targeting just ARv7, further showing that these genes are driven primarily by ARv7 (Figure 5F). In contrast, siRNA targeting just ARfl (exon 7 siRNA) had a much more modest effect on the ARv7-associated genes, while ARfl depletion with ARCC-32 modestly increased these genes.

Figure 5. ARv7 transcriptome in VCaP16 is highly correlated with ARfl in VCaP.

(A–E) Log₂(fold change) of significantly differentially expressed genes (p_adj < 0.05).

(F) Log₂(fold change) of ARv7-regulated genes (ARv7-binding sites ±5 kb from TSS) altered by siV7, siEx1, siEx7, or ARCC-32 in VCaP16. Differentially expressed genes with p_adj < 0.05 (siV7, siEx1, siEx7) and p < 0.05 (ARCC-32) were included. Mann-Whitney U test, two-tailed: siV7 versus siEx7, p = 0.0308; siV7 versus siEx1, p = 0.8151; siV7 versus ARCC-32, p <0.0001; siEx1 versus siEx7, p = 0.0609; ARCC-32 versus siEx7, p = 0.1148; ARCC-32 versus Ex1, p = 0.0005.

(G) Differentially expressed genes linked to ARv7-binding sites in VCaP16 cells (± 100 kb from TSS) in VCaP16 with siRNA knockdown of ARv7 (siV7) versus non-targeting control.

(H) Enrichment analysis by Enrichr of downregulated genes from (G).

The enhanced expression of many ARv7-stimulated genes, and the trend toward augmentation in the hallmark AR responsive gene set (see Figure 1H) following ARfl depletion, suggested that ENZ-liganded ARfl may function directly as a repressor. Alternatively, this may reflect coactivator proteins being sequestered by the ENZ-liganded ARfl, with these coactivators becoming available for ARv7 coactivation upon ARfl depletion. To test this hypothesis, we carried out ChIP-seq for H3K27ac, BRD4, and p300 in VCaP16 treated with vehicle or ARCC-32. ARCC-32 did not have a clear effect on H3K27Ac at ARv7- or ARfl-binding sites (Figure S8A). Moreover, it did not have an overall effect on BRD4 at ARv7- or ARfl-binding sites (Figure S8B). In contrast, p300 binding was decreased (Figure S8C). We then examined two directly ARv7-regulated genes whose expression was increased by ARCC-32, GRIN3A, and POLR2M (Figure S8D). For both genes, H3K27Ac and BRD4 at ARv7-binding sites was increased in response to ARCC-32, although p300 was decreased, possibly reflecting a greater role for CBP versus p300 as a coactivator for ARv7 (Figures S8E and S8F). These results suggest that coactivator redistribution may be one basis for effects of ARfl depletion.

Further examination of the genes with called ARv7-binding sites that were altered in response to ARv7 siRNA showed downregulation of multiple canonical AR-regulated genes such as KLK2 and KLK3 (Figure 5G) and of the hallmark androgen response gene set (Figure 5H). Gene sets related to mTORC1 signaling, tumor necrosis factor α signaling, and lipid metabolism were also significantly enriched, implicating ARv7 in restoration of these metabolic functions.

ARv7-binding sites are enriched for enhancer marks

The predominant role of ARv7 in driving the AR transcriptome suggested that ARv7-binding sites may have particularly strong enhancer activity. Therefore, we next carried out ChIP-seq for enhancer marks (H3K27ac and H3K4me1) in VCaP and VCaP16. Indeed, centering on all AR sites in VCaP16 cells, H3K27ac was higher at ARv7 versus ARfl-unique sites (Figure S9A). This same finding was observed in VCaP, indicating that the sites where ARv7 binds in VCaP16 are also potent enhancers in VCaP, which are presumably driven by the androgen-liganded ARfl. Notably, H3K27ac peak intensities across ARfl-unique and ARv7 sites were comparable in VCaP16 versus VCaP, consistent with the restoration of AR transcriptional activity in VCaP16. The results for H3K4me1 were similar to those for H3K27ac (Figure S9B). Together, these data indicate that ARv7 sites are enriched for enhancer activity but do not provide a basis for the gain in ARv7 binding to chromatin in VCaP16.

Adaptation to ENZ is associated with increased chromatin accessibility

We next found an increase in the intensity and number of assay for transposase-accessible chromatin (ATAC) sites in VCaP16 versus VCaP or VCaP-E cells (Figures 6A and 6B). We then identified the subset of ATAC sites that had significantly increased peak intensity in VCaP16 versus VCaP (ATAC-UP sites) (Figure 6C). Profile plots indicate that chromatin accessibility at the ATAC-UP sites was modestly increased after short-term ENZ (VCaP-E cells) and that there was also a modest overall intensity increase in sites identified as unchanged in VCaP16 versus VCaP (unchanged sites) (Figure 6D). Most ATAC sites were located in introns or intergenic, and comparison with RNA-seq data showed that ATAC-UP sites were correlated with increased expression of associated genes (Figure S10). Notably, ~40% of ARv7 sites overlapped ATAC-UP sites versus ~10% for ARfl-unique sites (Figure 6E). We then assessed ATAC-seq signal centered on ARv7-binding sites and found that this was increased in VCaP16 versus VCaP (Figure 6F). In contrast, the ATAC-seq signal at ARfl-unique sites was lower and much more modestly increased in VCaP16. These findings indicated that overall chromatin accessibility was increased in the VCaP16 cells, particularly at ARv7-binding sites.

Figure 6. Chromatin accessibility in VCaP16 cells is increased at ARv7-binding sites.

(A and B) (A) ATAC-seq signal and (B) numbers of ATAC sites in parental VCaP, VCaP with ENZ for 4 days (VCaP-E) and VCaP16.

(C and D) (C) Heatmap and (D) profile plots of ATAC-seq signal in VCaP, VCaP-E and VCaP16 cells. ATAC-seq signal centered at sites with increased ATAC signal (UP sites), sites called as unchanged (Unchanged sites) or sites with decreased ATAC signal (Down sites) in VCaP16 versus parental VCaP cells as identified with CoBRA pipeline.

(E) Overlap of differentially more accessible sites (ATAC-UP) with shared ARv7/ARfl (upper panel) and ARfl-unique binding sites (lower panel) in VCaP16.

(F and G) (F) ATAC-seq signal enrichment and (G) nucleosome occupancy by MNase-seq at ARv7/ARfl (upper panels) and ARfl-unique binding sites (lower panels) in VCaP16.

(H) Motif enrichment analysis by HOMER at ATAC-UP sites in VCaP16.

(I) NFI motif enrichment by HOMER at ATAC-seq differentially more accessible sites (UP), unchanged (Unchanged), or differentially closed (Down) sites in VCaP16 versus VCaP with CoBRA pipeline (****p < 0.0001, Fisher’s exact test).

(J) V-plot of NFI motif footprint at ATAC-UP sites in VCaP16 (centered on NFI motif).

The transposase used for ATAC-seq can generate shorter fragments in nucleosome-depleted areas. However, analysis of ATAC-seq fragment length did not reveal differences between ARv7 and ARfl sites in VCaP or VCaP16 (Figure S11). Therefore, we used micrococcal nuclease digestion with deep sequencing (MNase-seq) to identify nucleosome-protected versus -exposed regions. Notably, we found greater depletion of nucleosomes around ARv7 sites, but not ARfl sites, in VCaP16 (Figure 6G). Overall, these results indicate that VCaP cells adapt to ENZ through nucleosome depletion around ARv7 sites to enhance ARv7 binding.

We next examined the LuCaP series of PDXs⁴⁶ and identified models that had the greatest increases in ARv7 when they relapsed after castration.²² In parallel, we assessed enrichment for a gene set associated with ARv7,²² which together indicated that the castration-resistant model LuCaP 77CR had the highest ARv7 activity, followed by 105CR and 136CR (Figures S12A and S12B). Finally, we found that one of two LuCaP 77CR and 96CR tumors, and two of two 105CR and 136CR tumors, formed a distinct cluster by unsupervised clustering of ARv7-signature gene expression (Figure S12C). Based on these findings, we examined LuCaP 105 versus 105CR by ATAC-seq, which showed an overall increase in chromatin accessibility in LuCaP 105CR (Figure S13A). Moreover, ATAC-seq signals at ARfl sites identified in VCaP16 were also increased in the LuCaP 105CR versus parental 105 PDXs, and signals at ARv7 sites identified in VCaP16 were further markedly increased relative to the parental LuCaP 105 PDXs (Figure S13B). Together, these findings are consistent with similar adaptive mechanisms driving AR reactivation in this PDX.

Single-cell ATAC-seq suggests an intermediate stage in the transition to VCaP16

We also carried out single-cell ATAC-seq (scATAC-seq) on VCaP, VCaP-E, and VCaP16, which uncovered eight distinct clusters (Figure S14A). Cluster 1 was relatively specific to parental VCaP cells, while cluster 3 was relatively specific to VCaP16 cells. An intermediate cluster (cluster 2) was observed across all conditions, positioning itself between clusters 1 and 3. As with the bulk ATAC-seq, scATAC-seq demonstrated higher chromatin accessibility at AR motifs in VCaP16 versus VCaP (Figures S14B, S15A, and S15B). This heightened accessibility was particularly pronounced in the VCaP16 and VCaP-specific clusters (cluster 3 versus cluster 1) (Figures S14B and S15C). Notably, there was also increased accessibility at AR motifs in VCaP-E compared to VCaP cells (Figures S15A and S15B), indicating the initiation of chromatin changes following short-term exposure to ENZ.

Interestingly, cluster 2 exhibited chromatin features similar to those in cluster 3, suggesting that it may represent a pool of cells driving the development of ENZ resistance. Indeed, trajectory inference analysis showed a pseudotime progression from cluster 1 (VCaP-specific) through cluster 2 to cluster 3 (VCaP16-specific) (Figure S14C). A complementary analysis using another clustering algorithm revealed similar results (Figure S16).

ARv7 activity in VCaP16 is dependent on nuclear factor I transcription factors

The most highly enriched transcription factor motifs at the ATAC-UP sites in VCaP16 were the common AR/GR/PR/MR motifs (Figure 6H). As expected, there was also enrichment for FOXA1, GRHL2, and HOXB13 motifs. However, the second most enriched motifs were for nuclear factor I (NFI) family transcription factors (NFIA, NFIB, NFIC, NFIX). Comparison with ATAC-Unchanged and ATAC-Down sites similarly showed enrichment for the NFI motif at ATAC-UP sites (Figure 6I), and footprint analysis further supported enrichment of NFI transcription factors at ATAC-UP sites (Figure 6J).

We next assessed for enrichment of the androgen response element (ARE), FOXA1, HOXB13, and NFI motifs at ARv7 and ARfl-unique sites in VCaP-E versus VCaP16. Notably, while ARE and AR-half motifs were enriched at ARv7 versus ARfl-unique sites in both cells, there was greater enrichment of ARE motifs at ARv7 sites in the VCaP-E versus VCaP16, indicating that ARv7 binding was most highly biased toward canonical AREs after short-term ENZ treatment (Figures 7A and S17). Conversely, the FOXA1 motif was enriched at ARv7 sites in VCaP16 versus VCaP-E (Figure 7B), while enrichment for the HOXB13 motif was comparable at ARv7 sites in both cell lines (Figure 7C). In contrast, enrichment for the NFI motif was greater at ARfl-unique (which are weak ARv7 sites) versus called ARv7 sites in both VCaP-E and VCaP16, with greatest enrichment at ARfl-unique sites in VCaP-E (Figure 7D). One interpretation of this finding is that NFI proteins preferentially support ARfl and ARv7 binding at weaker AREs. Therefore, we next assessed the NFI motif at AR-binding sites with canonical versus non-canonical AREs and found higher enrichment at non-canonical AREs (Figure 7E).

Figure 7. ARv7 activity in VCaP16 is dependent on nuclear factor I transcription factors.

(A–D) Enrichment for the (A) ARE, (B) HOXB13, (C) FOXA1, and (D) NFI motifs by HOMER at ARv7/ARfl shared and ARfl-unique sites in VCaP after short-term ENZ treatment (VCaP-E) versus in VCaP16 cells.

(E) NFI motif enrichment by HOMER at AR-binding sites with canonical full AREs versus non-consensus AREs (others) in VCaP after short-term ENZ treatment (VCaP-E) versus in VCaP16.

(F and G) Pearson’s correlation of log₂(fold change) values of significantly differentially expressed genes (p_adj < 0.1) altered with siRNA knockdown of ARv7 (siV7) versus (F) siRNA knockdown of NFIB and NFIX (siNFIB/X) and versus (G) siRNA knockdown of FOXA1 (siFOXA1) in VCaP16.

(H) (Left) ARfl (AR-NT) and ARv7 in VCaP16 and (I) 22RV1 cells after knockdown with both NFIB and NFIX (siNFIB/X), and non-targeting control (siNTC) siRNA. (Right) Quantification of ARfl and ARv7 normalized to siNTC. Numbers indicate levels of ARfl and ARv7 reduction compared to siNTC.

(J) ARfl- and ARv7-binding sites based on ChIP-seq in VCaP16 with siRNA knockdown of NFIB and NFIX (siNFIB/X) versus non-targeted siRNA knockdown (siNTC).

(K) NFI motif enrichment by HOMER at AR-binding sites in normal prostate tissue (N) versus primary PC in two clinical datasets.

NFI proteins play major roles in the development of diverse tissues and can have oncogenic or tumor-suppressor functions.⁴⁷ Moreover, NFI proteins can bind nucleosomes, control chromatin loop boundaries, and increase chromatin accessibility.^48–52 Notably, previous studies in PC have identified interactions between AR, FOXA1, and NFI proteins, and in particular implicated NFIB in regulation of AR activity and PC development.^53–56 NFIB and NFIX are highly expressed in prostate (Figure S18A), with NFIX being most highly expressed among AR-expressing PC cell lines (Figure S18B). We confirmed by RT-qPCR the preferential expression of NFIX in LNCaP cells and of NFIX and NFIB in VCaP cells (Figure S18C). Consistent with this pattern of expression, DepMap CRISPR and RNAi screens showed dependency on NFIB and NFIX in AR-positive PC cells but not AR-negative PC cells (Figures S18D–S18F). Finally, our RNA-seq data also showed higher NFIB and NFIX (versus NFIA and NFIC) in VCaP and VCaP16, although the difference was more modest, and levels for each NFI were comparable in VCaP and VCaP16 cells (Figure S18G).

Based on these data, we assessed the effects of combined siRNA-mediated knockdown of NFIB and NFIX in VCaP16. RNA-seq confirmed that the siRNA decreased both NFIB and NFIX without compensatory increases in NFIA or NFIC (Figure S19A). This did not have a consistent effect on canonical AR-regulated genes, with two such genes (ACPP and NKX3.1, respectively) being significantly increased and decreased (Figure S19B) as well as a modest decrease in hallmark androgen response genes (Figure S19C). Hallmark gene sets related to cell cycle were more significantly decreased, while there was an increase in epithelial-to-mesenchymal transition genes (Figure S19C). To more directly assess the contribution of NFI to ARv7 function, we compared the transcriptional effects of NFIB/X versus ARv7 depletion. Notably, these were strongly correlated (Figure 7F). There was also a correlation between transcriptional effects of ARv7 and FOXA1 knockdown, but the correlation with NFIB/X depletion was greater (Figure 7G).

We next looked specifically at expression of genes linked to ARv7-binding sites versus ARfl-unique sites (which are weak ARv7 sites). There was a correlation between effects of ARv7 and NFIB/X knockdown at the former ARv7-binding site linked genes, and a stronger correlation with the latter ARfl-binding site linked genes, indicating that genes linked to these weaker sites are more dependent on NFIB/X (Figure S20A). A similar analysis for FOXA1 siRNA showed weaker correlations for both ARv7- and ARfl-unique site linked genes (Figure S20B). Interestingly, while the majority of Hallmark gene sets were decreased by ARv7 knockdown, many gene sets were increased by both FOXA1 and NFIB/X knockdown, suggesting cooperative effects of FOXA1 and NFI transcription factors at many genes that are not AR regulated (Figure S19C).

We next found that NFIB/X knockdown in VCaP16 decreased total ARfl and ARv7 protein in whole-cell lysates, but there was a markedly greater decrease in the chromatin-bound fraction for ARv7 (Figure 7H). We also examined 22Rv1 cells cultured in ENZ, which express high levels of ARv7, and similarly found that NFIB/X knockdown decreased ARv7 chromatin binding (Figure 7I). Finally, ChIP-seq in VCaP16 cells confirmed that NFIB/X depletion caused a marked decrease in ARv7-binding sites, with a much lesser effect on ARfl (Figure 7J).

These results indicate that NFI proteins in ENZ-adapted cells contribute to an increase in chromatin accessibility to support ARv7 and ARfl chromatin binding, particularly at weak AR-binding sites that do not have canonical AREs. Based on these observations, we examined AR-binding sites at genes that were altered by ARv7 siRNA in VCaP16. Notably, only a small fraction of AR-binding sites linked to these ARv7-regulated genes had canonical AREs (Figure S21A). We next similarly examined AR-binding sites at genes that were altered by both depletion of ARv7 and NFIB/X and found that most also had non-canonical AREs (Figure S21B). These findings further support the conclusion that NFI proteins enhance ARv7 activity at weak AREs.

Interestingly, previous studies have found that the AR cistrome in PC diverges from that in normal prostate epithelium.^43,44 Moreover, analysis of AR-binding sites in normal prostate versus primary PC shows that the sites in normal prostate are more highly enriched for canonical full AREs (Figure S22A). Notably, in two datasets we found that the NFI motif was significantly enriched at AR-binding sites in PC versus normal prostate (Figure 7K).^43,57 Furthermore, in both normal prostate and PC tissues, the NFI motif was enriched at AR-binding sites that did not have the canonical full ARE (Figures S22B and S22C). Together, these findings indicate that NFI proteins may also contribute to AR reprogramming during PC development. Consistent with this hypothesis, a previous study found that NFIB levels by immunohistochemistry (IHC) were increased in PC,⁵³ although analyses of NFIB and NFIX mRNA in multiple datasets do not show clear increases in tumor versus normal tissues (Figure S22D).

DISCUSSION

Increased ARv7 is associated with resistance to available ASI therapies, but the extent to which it drives the AR program remains to be determined. To model progression from androgen dependence to ENZ resistance, we generated ENZ-resistant VCaP16 cells. These cells had features observed most commonly in the clinic including restoration of the AR transcriptional program, high ARfl, and increased ARv7. Notably, depletion of ARv7, but not ARfl, markedly suppressed the AR transcriptional program. Consistent with these results, ARv7 complexes on chromatin were not dependent on or associated with ARfl. Significantly, while ARv7 was increased acutely in response to ENZ (VCaP-E cells), we found that further adaptations were needed to enhance ARv7 binding to chromatin and its ability to drive the AR transcriptional program.

The gain in ARv7 chromatin binding in VCaP16 was associated with a broad increase in chromatin accessibility. This increase was greater at identified ARv7-binding sites relative to ARfl-unique sites (where ARv7 binding is below the peak calling threshold), and the depletion of nucleosomes at ARv7-binding sites in VCaP16 was confirmed by MNase-seq. Motifs enriched at sites with increased chromatin accessibility in VCaP16 (ATAC-UP sites) included those for AR, FOXA1, and HOXB13, but the most enriched motif (after the AR motif) was the NFI motif. Moreover, the transcriptional effects of depleting NFIB/X in VCaP16 were strongly correlated with those of ARv7 depletion, and combined knockdown of NFIB/X markedly decreased ARv7 chromatin binding. Together, these findings indicate a major role for NFI proteins in the enhancement of chromatin accessibility and in supporting ARv7 function.

Previous studies have shown varying overlap between the ARv7 versus ARfl cistromes and transcriptomes and indicated that ARv7 may have distinct functions.^26,38–42 However, in LNCaP95 (expressing endogenous ARfl and ARv7), we reported previously that all ARv7-binding sites were also ARfl sites.²⁷ We similarly found in VCaP16 that the vast majority of ARv7 sites were also ARfl sites. While these results indicate that the ARv7 cistrome largely mirrors that of ARfl, this does not rule out ARv7-unique sites in some contexts. Indeed, ZFX has been reported to be a novel ARv7 partner that drives ARv7 binding to unique non-canonical sites.⁵⁸ High levels of ARv7, either endogenous or exogenous, may possibly also drive ARv7 to unique sites in some contexts.⁵⁹

While the number of called ARv7 sites was relatively low, peak normalized counts for ARv7 were highly correlated with those for ARfl, and this correlation extended to ARfl peaks that were not called as ARv7 sites. These data indicate that the ARv7 cistrome largely overlaps that of ARfl, but with ARv7 having lower affinity and hence ARv7-binding sites being undercalled. Indeed, previous fluorescence recovery after photobleaching studies have shown that nuclear ARv7 has markedly higher mobility relative to agonist liganded ARfl, indicating a more rapid DNA-binding off rate.^33,35 We propose this relatively weak ARv7 chromatin-binding capacity is why adaptations that enhance chromatin accessibility are needed to support its activity in VCaP16 cells.

With respect to the ARv7 transcriptome, we found a strong correlation between genes regulated by ARv7 in VCaP16 cells and those regulated by ARfl in VCaP, supporting the conclusion that ARv7 takes over the transcriptional functions of ARfl. In contrast, there was minimal correlation between effects of depleting ARv7 versus ARfl in VCaP16 cells. Indeed, ARfl degradation increased expression of many genes that were decreased by ARv7 knockdown. This may reflect a direct repressive effect of ENZ-liganded ARfl and/or increased availability of some coactivator proteins that are being non-productively engaged by the ENZ-liganded ARfl. Significantly, either mechanism could mitigate beneficial effects of ARfl degraders that are now being tested in the clinic.

Previous studies have found that ARv7 sites are associated with open chromatin, with consensus full AREs, and with HOXB13.^40,41 Notably, HOXB13, but not FOXA1, has been found to bind directly to ARv7, consistent with an increased dependence on HOXB13 versus FOXA1.^40,41,60,61 We similarly found that ARv7-binding sites in VCaP16 were enriched for consensus full AREs and for the HOXB13 motif but also for the FOXA1 motif. Interestingly, consensus ARE enrichment was greatest in VCaP-E cells, indicating that increased chromatin accessibility in the VCaP16 was allowing ARv7 to bind to lower-affinity AREs. Notably, in contrast to HOXB13 and FOXA1 motifs, the NFI motif was enriched at ARfl-unique versus ARv7 sites, suggesting a particularly important role in supporting ARv7 binding at weak AREs.

NFI proteins bind DNA as homo- or heterodimers and can act as pioneer factors by binding to nucleosomes and increasing chromatin accessibility.^48–52 In addition to roles in the development of diverse tissues, they have context-dependent oncogenic or tumor-suppressor functions.^47,48,52,62 Previous studies in PC cells have found an association of NFI proteins with AR-binding sites^54–56 and identified direct interactions between NFI proteins and FOXA1 that may stabilize an AR-FOXA1-NFI complex.⁵⁵ In functional studies, NFIB knockdown in LNCaP cells had variable effects on AR-regulated genes, while prostate tissue explants from Nfib gene knockout mice showed hyperplasia that may reflect an AR-independent effect.⁵⁴ Our findings are consistent with these previously described interactions between NFI proteins and AR and reveal a further major role for NFI proteins in the progression to ENZ resistance.

We further hypothesize a role for NFI proteins in AR reprogramming during PC development,⁴³ as the NFI motif is significantly enriched at AR-binding sites in PC versus normal prostate epithelium. Notably, based on mRNA levels, NFI expression is not increased during PC development or the progression to ENZ resistance, although one IHC study found increased NFIB protein in primary PC.⁵³ However, NFI protein levels or activity may be modulated by alternative splicing or post-translational mechanisms.⁶³ Further studies to address NFI expression and activity during PC development and progression are warranted.

Limitations of the study

We have not completely eliminated expression of ARfl in VCaP16 cells and so cannot rule out some role for the remaining low level of ARfl. Conclusions are based primarily on the transition to ENZ resistance in VCaP cells, so there may be mechanisms that can activate the ENZ-liganded ARfl in some patients, including low levels of residual androgens that are not effectively blocked by AR antagonists. While we find that the ARv7 cistrome largely overlaps that of ARfl, other studies have found novel functions for ARv7. Further studies are clearly needed to further address the extent to which ARv7 (or related variants) drive the AR program in vivo in patients and to identify novel dependencies and functions, which may be context dependent.

RESOURCE AVAILABILITY

Lead contact

Further information can be obtained through the lead contact, Steven Balk (sbalk@bidmc.harvard.edu).

Materials availability

Cell lines generated can be obtained through the lead contact.

Data and code availability

• The public datasets analyzed in this paper are in the Gene Expression Omnibus (GEO), and their accession numbers are listed in the key resources table.

• New genomic data generated here have been deposited in the GEO and are available as of the date of publication. Accession numbers are listed in the key resources table.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
anti-Vinculin	Santa Cruz Biotechnology	Cat#sc-73614; RRID:AB_1131294
anti-GAPDH	Cell Signaling Technologies	Cat#5174S; RRID:AB_10622025
anti-AR (NT)	RevMab Biosciences	Cat#31-1135-00; RRID:AB_2716455
anti-ARv7	RevMAb Biosciences	Cat#31-1109-00; RRID:AB_2716436
anti-CBP-P300	Thermo-Fisher	Cat#MA5-13634; RRID:AB_10987120
anti-FOXA1	Cell Signaling Technologies	Cat#53528S; RRID:AB_2799438
anti-FOXA1	Sigma	Cat#WH0003169M1; RRID:AB_1841663
anti-PSA	Cell Signaling Technologies	Cat#5365S; RRID:AB_2797609
anti-H3	Cell Signaling Technologies	Cat#4499S; RRID:AB_10544537
anti-AR (C-terminal)	Abcam	cat#AB227678; RRID:AB_2833098
anti-NFIB	Bethyl	cat#A303-566A; RRID:AB_10951938
anti-NFIB	Sigma	cat#HPA003956; RRID:AB_1854424
anti-NFIX	Thermo Fisher	cat#PA5-64917; RRID:AB_2663319
anti-H3K27Ac	Diagenode	cat#C15410196; RRID:AB_2637079
anti-H3K4me1	Abcam	cat#ab8895; RRID:AB_306847
anti-BRD4	Diagenode	cat#C15410337; no RRID
anti-p300	Cell Signaling Technologies	cat#57625S; RRID:AB_3068009
Biological samples
VCaP cell line	ATCC	RRID:CVCL_2235
LNCaP-95 cell line	Myles Brown Lab	RRID:CVCL_ZC87
Chemicals, peptides, and recombinant proteins
Enzalutamide	SelleckChem	Cat#S1250
ARCC-32	Arvinus Pharmaceuitcals	Contact Arvinus Pharmaceuitcals
RIPA	Fisher	Cat#89901
Triton X-100	Sigma-Aldrich	Cat#93443-100 mL
RNaseA	Thermo Fisher Scientific	Cat#EN0531
Dynabeads (Protein A)	Thermo Fisher Scientific	Cat#10002D
Dynabeads (Protein G)	Thermo Fisher Scientific	Cat#10004D
Paraformaldehyde	Thermo Fisher Scientific	Cat#128800
PBS	Thermo Fisher Scientific	Cat#MT21040CV
TBS	Thermo Fisher Scientific	Cat#J62938.K7
Hoechst 33342	Thermo Fisher Scientific	Cat#H3570
Prolong Gold anti-fade mounting media	Thermo Fisher Scientific	Cat#P36931
TaqMan Fast Virus 1-Step Master Mix	Thermo Fisher Scientific	Cat#4444434
GAPDH TaqMan Primer Probe	Thermo Fisher Scientific	Cat#4310884E
AR-FL TaqMan Primer Probe	Thermo Fisher Scientific	Cat#Hs001711772_m1
AR-v7 TaqMan Primer Probe	Thermo Fisher Scientific	Cat#4331348, Custom Assay ID: AI6ROCI
KLK3 TaqMan Primer Probe	Thermo Fisher Scientific	Cat#Hs02576345_m1
KLK2 TaqMan Primer Probe	Thermo Fisher Scientific	Cat#Hs00428383_m1
FKBP5 TaqMan Primer Probe	Thermo Fisher Scientific	Cat#Hs01561006_m1
TMPRSS2 TaqMan Primer Probe	Thermo Fisher Scientific	Cat#Hs01120965_m1)
SLC45A3 TaqMan Primer Probe	Thermo Fisher Scientific	Cat#Hs01026319_g1
Dexamethasone	Sigma	Cat#D4902
ON-TARGETplus Human NFIB siRNA, SMARTPool	Horizon	Cat#L-008456-00-0005
ON-TARGETplus Human NFIX siRNA, SMARTPool	Horizon	Cat#L-009250-00-0005
ON-TARGETplus Human FOXA1 siRNA, SMARTPool	Horizon	Cat#L-010319-00-0005
ON-TARGETplus Non-targeting Control Pool	Horizon	Cat#D-001810-10-05
Disuccinimidyl glutarate	Thermo Fisher Scientific	Cat#20593
MNase	Worthington	Cat#LS004797
Exonuclease III (E. coli)	New England BioLab	Cat#M0206S
Proteinase K	Thermo Fisher Scientific	Cat#EO0491
DSP	Thermo Fisher Scientific	Cat#22585
IP buffer	Thermo Fisher Scientific	Cat##87788
Benzonase	Sigma	Cat#71205
2X tagmentation buffer	Illumina	Cat#70615027866
Tn5 transposase	Illumina	Cat#70715027865
Critical commercial assays
ThruPLEX DNA-seq Kit	Rubicon Genomics	N/A
Accel-NGS 2S Plus DNA Libraries Kit with Unique Dual Indexing	Swift Biosciences	cat# 29096/290384
SimpleChIP ChIP-seq Multiplex Oligos for Illumina with Dual Index Primers	Cell Signaling Technology	cat# 47538
ChIP DNA Clean & Concentrator™ Kit	Zymo Research	cat# D5205
Blood & Cell Culture DNA Kit Midi Kit	Qiagen	cat#13343
Qubit dsDNA High Sensitivity Kit	Thermo Fisher Scientific	cat#Q32854
RNeasy Plus Mini Kit	Qiagen	cat#74136
QIAquick PCR Purification Kit	Qiagen	cat#28104
Agilent High Sensitivity DNA Kit	Agilent	cat#5067-4626
Agilent High Sensitivity D1000 ScreenTape System	Agilent	cat#5067-5584/5067-5585
Chromium Next GEM Single Cell ATAC Library & Gel Bead Kit	10X Genomics	Cat#PN-1000176
MycoAlert detection kit	Lonza	Cat#LT07-418
BCA Protein Assay Kit	Thermo Fisher Scientific	Cat#23225
Subcellular Protein Fractionation Kit	Thermo Fisher Scientific	Cat#78840
CyQUANT™ Direct Cell Proliferation Assay	Thermo Fisher Scientific	Cat#C35011
Deposited data
AR ChIP-seq (public)	Gene Expression Omnibus	GSE106561 (GSM2842708, GSM2842700, GSM2842704), GSE99378 (GSM2643239, GSM2643240, 779 GSM2643241, GSM2643242, GSM2643245, GSM2643246, GSM2643247, GSM2643248), GSE143906 780 (GSM4276560, GSM4276561), GSE130408 (GSM4569829 - GSM4569867), GSE39879 (GSM980655 781 GSM980664), GSE137527 (GSM4081278 - GSM4081325)
RNA-seq data (public)	Gene Expression Omnibus	GSE126078
Expression profiling by array (public)	Gene Expression Omnibus	GSE21034 (GSM526290 - GSM5262318, GSM526265 - GSM526283, GSM526134 - GSM526264)
ChIP-Seq (this study)	Gene Expression Omnibus	GSE252897
scATAC-seq (this study)	Gene Expression Omnibus	GSE252843
WES (this study)	Gene Expression Omnibus	GSE252842
ATAC-seq (this study)	Gene Expression Omnibus	GSE252840
ChIP-Seq (this study)	Gene Expression Omnibus	GSE277606
Oligonucleotides
siRNA ARv7 (Target sequence: GUAGUUGUAUCAUGAUU)	Dharmacon	Cat#SRYWD-000001
siRNA AR Exon7 (Target sequence: UCAAGGAACUCGAUCGUAUUU)	Dharmacon	Cat#VOZOB-000001
siRNA AR Exon1 (Target sequence: CAAGGGAGGUUACACCAAAUU)	Dharmacon	Cat#SRYWD-000005
FKBP5_1_F (CTC TTG CTG GAA CAC GAG GT)	Integrated DNA Technologies	Manufacturing ID# 400391426
FKBP5_1_R (TGC CCT CCT ATG GGC TAC TT)	Integrated DNA Technologies	Manufacturing ID# 400391427
FKBP5_2_F (CTG CGC AAT CGG AGT GTA AC)	Integrated DNA Technologies	Manufacturing ID# 400391428
FKBP5_2_R (ATC GAG TTC ATG TGC CAG CC)	Integrated DNA Technologies	Manufacturing ID# 400391429
SCNN1A_F (GGG CCC TAG GAC ATT CTG TT)	Integrated DNA Technologies	Manufacturing ID# 400391434
SCNN1A_R (TTC CTG CAA CTC TGT GAC CA)	Integrated DNA Technologies	Manufacturing ID# 400391435
Software and algorithms
Visualization Pipeline for RNA-seq analysis (VIPER)	Cornwell et al.64	NA
CHromatin enrichment ProcesSor Pipeline (CHIPS)	Taing et al.65	NA
Genome Analysis Toolkit (GATK 4.1.9.0) Best Practices Workflows	Genome Analysis Toolkit	NA
Cell Ranger ATAC pipeline, version 1.2.0	10X Genomics (https://support.10xgenomics.com/single-cell-atac/software)	NA
STAR aligner (2.7.0f)	Dobin et al.66	NA
Cufflinks (v2.2.1)	Trapnell et al.67	NA
BEDTools (v2.27.1)	Quinlan and Hall68	NA
Burrows-Wheeler Aligner 0.7.15-r1140	Li and Durbin69	NA
MACS2 algorithm (2.2.7.1)	Zhang et al.70	NA
HOMER v3.12	Heinz et al.71	NA
Samtools 1.9	Danecek et al.72	NA
deepTools 3.0.2	Ramírez et al.73	NA
Bamliquidator	Documentation:https://github.com/BradnerLab/pipeline/wiki/bamliquidator	NA
Gene Set Enrichment Analysis (GSEA)	https://www.gsea-msigdb.org/gsea/index.jsp	NA
EnrichR	http://amp.pharm.mssm.edu/Enrichr	NA
Containerized Bioinformatics workflow for Reproducible ChIP/ATAC-seq Analysis (CoBRA)	https://bitbucket.org/cfce/cobra; cfce-cobra.readthedocs.io/en/latest/	NA
Loupe Browser, version 6.5.0	10X Genomics	NA
Slingshot algorithm (Trajectory Inference for Single-Cell Data)	https://bioconductor.org/packages/devel/bioc/vignettes/slingshot/inst/doc/vignette.html	NA
Seurat	https://satijalab.org/seurat/	NA
Signac	https://stuartlab.org/signac/

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Cell culture

VCaP cells (ATCC, Manassas, VA) were maintained in DMEM with 10% fetal bovine serum (FBS). LNCaP95 cells⁷⁴ were maintained in RPMI 1640 medium supplemented with 10% charcoal/dextran-stripped fetal bovine serum (CSS), 1% L-Glutamine and 1% penicillin/streptomycin. Cell line identities were established by short tandem repeat profiling and interspecies contamination test (IDEXX). Cells were screened for mycoplasma with MycoAlert detection kit (Lonza) at receipt or thaw, and then monthly for the duration of project. Cells were employed for 20 passages only before new cell lines were thawed. For generation of VCaP16 ENZ-resistant model, we seeded cells on plates in standard parental media and, after overnight attachment, switched to medium with 2-fold increasing concentrations of ENZ, ending at 16 μM ENZ after ~2 months. The VCaP16 cells were maintained in 16 μM ENZ, and medium on all cell models was changed every 2–3 days.

METHOD DETAILS

Immunoblotting and immunoprecipitation

For immunoblotting cells were harvested in RIPA lysis buffer supplemented with a protease and phosphatase inhibitor cocktails (Thermo Fisher). Lysis was allowed to proceed for 30–60 min on ice, before cells were spun down at maximum RPM in a microfuge and cell debris pellets removed. Protein concentrations were determined by Bicinchoninic acid (BCA) assay (Thermo Fisher). Lysates were prepared with 4X Laemmli buffer (BioRad, #1610747) and β-mercaptoethanol. Samples were run on pre-cast 4–15% gradient gels (BioRad), and fast-transferred to nitrocellulose membranes using the TransBlot Turbo system (BioRad). Separation of cells into cytoplasmic, soluble nuclear, and chromatin fractions was carried out using a Subcellular Protein Fractionation Kit (Thermo Fisher #78840).

For assessment of ARv7/ARfl interaction, cells were incubated in PBS with 1.5 mM dithiobis (succinimidyl propionate) (DSP, Thermo Fisher #22585) for 20 min, followed by incubation in 1% formaldehyde for 10 min at room temperature. Cross-linking was quenched by adding glycine to 0.1 M. Cell nuclei were then extracted and resuspended in 1/2 dilution IP buffer (Thermo Fisher #87788) and sonicated for 2 min (5s on/5s off) in a Diagenode Bioruptor in 4°C water bath with medium power setting. The sample was combined with 10×TBS and 10% Triton X-100 to give final concentration of 1X TBS/1%Triton X-100, and 100 mM MgCl2 was added to give 1 mM MgCl2 final concentration. Benzonase (Sigma #71205) was then added (90 U per 50 μL, which was determined to be optimum in initial studies).³¹ Benzonase digestion was for 15 min at 37°C on a 750 RPM shaker, and then 50 μL of each aliquot was processed for reverse cross-linking, followed by immunoblotting.

Immunofluorescence

Cells were fixed with 4% paraformaldehyde at room temperature on a rocker in the dark, and then washed three times with PBS. Fixed cells were washed with TBS and permeabilized with 0.1% Triton X-100 for 10 min at room temperature. After three washes with TBS, the sections were incubated with 5% normal donkey serum (Jackson ImmunoResearch Lab Inc, West Grove, PA) for an hour at room temperature. Slides were then incubated with mouse anti-Androgen Receptor C-terminal (ARfl) (1:100, Abcam #ab227678), rabbit anti-ARv7 (1:200, RevMab Biosciences, 31–1109-00), and mouse anti-CBP/p300 monoclonal antibody (NM11, Thermo-Fisher MA5–13634) overnight at 4°C. The slides were washed three times and incubated with Alexa Fluor 647 conjugated donkey anti-mouse secondary antibody (Jackson ImmunoResearch Lab, 1:300) or Cy3 conjugated donkey anti-rabbit secondary antibody (Jackson ImmunoResearch Lab, 1:300). Samples were counterstained with Hoechst 33342 (Invitrogen) and washed three times with TBS. The slides were mounted with Prolong Gold anti-fade mounting media (Invitrogen). A Zeiss LSM 880 Inverted Live-cell Laser Scanning Confocal Microscope with Airyscan module was employed for all imaging. z stack images were captured for all markers to generate a 3D dataset for quantification with Imaris 9.8.2 software (Oxford Instruments).

Quantitative real-time PCR

RNA was extracted from cell lines using an RNeasy Plus kit (Qiagen) according to the manufacturer’s instructions. Subsequent qPCR was performed using the TaqMan Fast Virus 1-Step Master Mix and StepOnePlus Real-Time PCR System (Thermo Fisher). The primers purchased from TaqMan were: GAPDH (Thermo Fisher, #4310884E), AR-FL (Thermo Fisher, Hs001711772_m1), AR-v7 (Thermo Fisher, #4331348, Custom Assay ID: AI6ROCI), KLK3 (Thermo Fisher, Hs02576345_m1), KLK2 (Thermo Fisher, Hs00428383_m1), FKBP5 (Thermo Fisher, Hs01561006_m1), TMPRSS2 (Thermo Fisher, hs01120965_m1), and SLC45A3 (Thermo Fisher, Hs01026319_g1).

Proliferation assays

For proliferation of VCaP cells exposed to ENZ, cells were plated at 20,000 cells/well on day - 2. Experimental and control treatments were conducted and initial readings taken on day 0. CyQuant Direct Cell Proliferation Assay (Invitrogen) was employed for determination of cell number based on fluorescent DNA staining, with dead cell exclusion provided by CyQuant Suppressor solution (Invitrogen). An equal volume of CyQuant solution, prepared on treatment day, was added to cells in culture medium, incubated for 60 min at 37°C, and fluorescence read on a SpectraMAX iD3 microplate reader (excitation/emission 485 nm/525 nm). To assess effects of depleting ARfl or ARv7, cells were treated with siARv7, ARCC-32 (obtained from Arvinas, New Haven, CT), or vehicle on days 0 and 4.

siRNA transfection

Transfection with RNAiMAX reagents (Invitrogen) was conducted per manufacturer’s instructions for up to 72 h. The siRNA for non-targeting control (Dharmacon, #D-001810–01-05), ARfl (exon 7, Dharmacon, #VOZOB-000001, UCAAGGAACUCGAUCGUAUUU), ARv7 (Dharmacon, #SRYWD-000001, GUAGUUGUAUCAUGAUU) or both (exon 1, Dharmacon, #SRYWD-000005, CAAGGGAGGUUACACCAAAUU) were purchased from Dharmacon.

Next generation sequencing

Cell Culture Conditions: VCaP and VCaP16 cells were cultured in complete DMEM media supplemented with 10% FBS and 1% Penicillin (5000 I.U/ml)-Streptomycin (5000 μg/mL) solution. VCaP16 cells were cultured with the addition of 16 μM ENZ. Twenty-four hours prior to harvesting, a new culture medium was prepared and replaced. Genomic DNA was isolated using DNeasy Blood & Tissue Kit (Qiagen). Libraries for Whole Exome Sequencing (WES) were prepared, and the sequencing was performed by Novogene, located in Sacramento, CA. The NovaSeq6000 PE150 platform was employed for this purpose.

RNA-sequencing (RNA-seq)

Total RNA was isolated, mRNA was sent for libraries generation by the Center for Functional Cancer Epigenetics (Dana-Farber Cancer Institute (DFCI)) using the Illumina TruSeq stranded mRNA sample kit followed by sequencing on the Illumina NextSeq500 platform at the Molecular Biology Core Facility (DFCI); or mRNA was sent to Novagene. (Sacramento, CA) for libraries preparation and sequencing on NovaSeq6000 platform.

ChIP-sequencing

ChIP-seq was performed as previously described.²⁷ The cells were cultured and plated to ensure they reached a healthy confluency of 90% on 10 cm plates before crosslinking. Briefly, cells were crosslinked with 1% formaldehyde (single fixation) or 2 mM fresh disuccinimidyl glutarate and 1% formaldehyde (double fixation) and then chromatin sonicated to 200–600 bp. ChIP was carried out using antibodies against H3K27Ac (Diagenode, C15410196), H3K4me1 (Abcam, ab8895), C-terminal ARfl (Spring Bioscience, clone SP242), ARv7 (RevMAb Biosciences USA Inc., 31–1109-00), FOXA1 (Cell Signaling Technologies, 53528S), BRD4 (Diagenode, C15410337), p300 (Cell Signaling Technologies, 57625S) and then conjugation to Protein A and G Dynabeads (Thermo Fisher Scientific). ChIP DNA was purified using the ChIP DNA Clean & Concentrator Kit (Zymo Research, D5205). ChIP-seq libraries were generated using the ThruPLEX DNA-seq Kit (Rubicon Genomics), Accel-NGS 2S Plus DNA Libraries Kit with Unique Dual Indexing (Swift Biosciences, 29096/290384) or SimpleChIP ChIP-seq Multiplex Oligos for Illumina with Dual Index Primers (Cell Signaling Technology, 47538) according to the manufacturer’s instructions. Libraries were pooled and sequenced on the Illumina NextSeq500 platform at the Molecular Biology Core Facility (DFCI) or NovaSeq6000 platform at the Novogene (Sacramento, CA).

MNase-sequencing

MNase-seq was performed as previously described.⁷⁵ VCaP and VCaP16 cells were cultured in complete DMEM media supplemented with 10% FBS and 1% Penicillin (5000 IU/mL)-Streptomycin (5000 μg/mL) solution. VCaP16 cells were cultured with the addition of 16 μM ENZ. Twenty-four hours prior to harvesting, fresh culture medium was added. Cells were crosslinked with 1% formaldehyde and then 2.5 × 10⁶ cells were treated with 60 U of MNase (Worthington, LS004797) and Exonuclease III (E. coli) (New England BioLab, M0206S). DNA was purified using ChIP DNA Clean & Concentrator Kit (Zymo Research, D5205). Libraries were generated using the SimpleChIP ChIP-seq Multiplex Oligos for Illumina with Dual Index Primers (Cell Signaling Technology, 47538) according to the manufacturer’s instructions. Libraries were pooled and sequenced on the NovaSeq6000 platform at the Novogene (Sacramento, CA).

ATAC-seq (bulk & single-cell)

For bulk ATAC-seq cells were crosslinked with fresh 1% formaldehyde for 10 min, then quenched with glycine and frozen. Cells were thawed in cold ATAC-seq buffer (ASB, 10 mM Tris-HCl pH 7.4, 10 mM NaCl, and 3 mM MgCl₂) and centrifuged. Cells were lysed by resuspension in 0.5 mL of lysis buffer (ASB with 0.1% NP40, 0.1% Tween 20, and 0.01% digitonin), followed by addition of 1 mL ASB with 0.1% Tween 20 and centrifugation. Nuclei were resuspended in 50 μL tagmentation reaction containing 25 μL 2X tagmentation buffer (Illumina 15027866), 16.5 μL PBS, 0.5 μL Tween 20, 0.5 μL 1:1 water diluted digitonin, 2.5 μL Tn5 transposase (Illumina 15027865), and 5 μL water, and incubated at 37°C for 30 min. Transposed DNA was purified using Qiagen columns. Libraries were amplified and sequencing was performed in thirty-five base pair paired-end reads on a NextSeq instrument (Illumina). For single cell ATAC-seq, approximately 7000 cells were targeted for each sample and processed according to the 10X Genomics scATAC-seq sample preparation protocol (Chromium Single Cell ATAC Library & Gel Bead Kit, 10X Genomics). Libraries were sequenced as paired-end reads on a NextSeq instrument (Illumina).

RNA-seq analysis

Total RNA was isolated using the RNeasy Plus Kit (Qiagen) following the manufacturer’s instructions. mRNA libraries were generated using the Illumina TruSeq stranded mRNA sample kit. Raw reads were analyzed using a pipeline for RNA-seq analysis - Visualization Pipeline for RNA-seq analysis (VIPER) based on workflow management system Snakemake (https://github.com/hanfeisun/viperrnaseq). The read alignment to hg19 reference genome was performed using STAR aligner (2.7.0f) with default parameters. Gene expression (FPKM values) was quantitated with Cufflinks (v2.2.1). Bedtools (v2.27.1) genomecov and bedGraphToBigWig v 4 were used to generate bigwig files.

ChIP-seq analysis

The data were analyzed using a pipeline CHromatin enrIchment ProcesSor Pipeline (CHIPS) based on workflow management system Snakemake (https://github.com/liulab-dfci/CHIPS) unless otherwise stated. Raw reads were aligned to hg19 reference genome using Burrows-Wheeler Aligner (bwa mem) 0.7.15-r1140 with default parameters; MACS2 algorithm (2.2.7.1) was used to call peaks; MACS2 (2.2.7.1) and bedGraphToBigWig (v 4) were used to generate bigwig files. Read counts for downstream analyses were quantified using the bioliquidator/bamliquidator tool from Docker image (https://hub.docker.com/r/bioliquidator/bamliquidator). Genomic regions were annotated using Hypergeometric Optimization of Motif EnRichment tool (HOMER v3.12). Motif enrichment analysis was performed using HOMER v3.12. Output bam files (from CHIPS pipeline) for ARv7 and ARfl in VCaP and VCaP16 cells were used for secondary peak calling with HOMER v3.12 to generate bed files that were used for further downstream analysis. Bigwig files for samples with biological duplicates were averaged using a combination of the wiggletools and wigToBigWig tools. Additionally, a custom in-house pipeline was employed for the analysis of ChIP-seq data. FastQC 0.11.5 and MultiQC 1.9 tools were used for quality control assessment of fastq files. Then, the paired end reads were trimmed with Trimmomatic 0.36 tool. Trimmed paired end reads were mapped to hg19 with Burrows-Wheeler Aligner 0.7.8 (bwa mem -M -t 4). Aligned reads in a generated sam file were sorted by coordinates with SamSort function from Picard 2.8.0 tool with following parameters SORT_ORDER = coordinate VALIDATION_STRINGENCY = SILENT. Duplicated reads were marked with MarkDuplicates function from Picard 2.8.0 tool with VALIDATION_STRINGENCY = SILENT ASSUME_SORTED = true. The ENCODE blacklist regions were removed with bedtools 2.27.1. Duplicated reads were removed with MarkDuplicates function from Picard 2.8.0 tool with following flags ASSUME_SORTED = true VALIDATION_STRINGENCY = SILENT REMOVE_DUPLICATES = true. Unmapped reads, not primary reads, reads aligned to different locations and low-quality reads were filtered out with samtools 1.9 using following parameters samtools view -F 0×104 -b my.bam -q 30. Indexing of bam files was done with samtools 1.9. BibWig files were generated with bamCoverage function from deeptools 3.0.2 with following parameters –binSize 10 –normalizeUsing BPM –smothLenght 60 –extendReads 150 –centerReads. Peaks were called using MACS2 2.1.1. algorithm with default parameters. Profile plots and heatmaps were generated with deeptools 3.0.2.

MNase-seq analysis

The raw sequencing data underwent analysis using a modified version of the nf-core/mnaseseq pipeline. The paired end reads were trimmed with Trimmomatic 0.36 tool using following parameters ILLUMINACLIP:TruSeq3-PE.fa:2:30:10:2:TRUE LEADING:3 TRAILING:3 MINLEN:20 (considering short< 50 bp fragments). Trimmed paired end reads were mapped to hg19 with Burrows-Wheeler Aligner 0.7.8 (bwa mem -M -t 4). Aligned reads in a generated sam file were sorted by coordinates with SamSort function from Picard 2.8.0 tool with following parameters SORT_ORDER = coordinate VALIDATION_STRINGENCY = SILENT. Duplicated reads were marked with MarkDuplicates function from Picard 2.8.0 tool with VALIDATION_STRINGENCY = SILENT ASSUME_SORTED = true. The ENCODE blacklist regions were removed with bedtools 2.27.1. Duplicated reads were removed with MarkDuplicates function from Picard 2.8.0 tool with following flags ASSUME_SORTED = true VALIDATION_STRINGENCY = SILENT REMOVE_DUPLICATES = true. Unmapped reads, not primary reads, reads aligned to different locations and low-quality reads were filtered out with samtools 1.9 using following parameters samtools view -F 0×104 -b my.bam -q 30. Indexing of bam files was done with samtools 1.9. Bigwig files were generated with bamCoverage function from deeptools 3.0.2 with following parameters –binSize 10 –normalizeUsing BPM –smothLenght 60 –extendReads 150 –centerReads. Profile plots and heatmaps were generated with deeptools 3.0.2 to visualize the nucleosome footprints.

WES analysis

Genomic DNA was isolated using DNeasy Blood & Tissue Kit (Qiagen). Libraries for Whole Exome Sequencing (WES) were prepared, and the sequencing was performed on NovaSeq6000 platform at the Novogene (Sacramento, CA) aiming for 100X coverage. The analysis of WES data followed the established Genome Analysis Toolkit (GATK 4.1.9.0) Best Practices Workflows. The raw reads were aligned to the hg38 reference genome using the Burrows-Wheeler Aligner (BWA) version 0.7.8. The identification of candidate somatic variants was carried out using the Mutect2 function from GATK version 4.1.9.0. Identified somatic variants were annotated using Funcotator, which is a part of GATK 4.1.9.0. ANNOVAR (ANNOtate VARiation) tool was utilized for the interpretation and prioritization of the filtered somatic variants. To assess copy ratio alterations, we employed a suite of functions from GATK 4.1.9.0, including PreprocessIntervals, CollectReadCounts, DenoiseReadCounts, and PlotDenoisedCopyRatios, which also allowed for visualization of the data.

ATAC-seq analysis

The data were analyzed using a pipeline CHromatin enrIchment ProcesSor Pipeline (CHIPS) based on workflow management system Snakemake (https://github.com/liulab-dfci/CHIPS) and Containerized Bioinformatics workflow for Reproducible ChIP/ATAC-seq Analysis (CoBRA) (https://bitbucket.org/cfce/cobra;cfce-cobra.readthedocs.io/en/latest/ ).

V-plot

V-plot of differentially more accessible sites, ATAC-UP sites, centered on NFI motif. Each dot represents the midpoint of each paired-end fragment placed on the graph. Y axis represents the length and X axis represents the distance of the paired-end fragment midpoint from the center of NFI motif. The left diagonal results from fragments cut by Tn5 transposase precisely to the right of the transcription factor-protected region, and the converse is true of the right diagonal.

scATAC-seq analysis

The raw FASTQ files underwent preprocessing using the Cell Ranger ATAC pipeline, version 1.2.0, provided by 10X Genomics (https://support.10xgenomics.com/single-cell-atac/software). The data was aligned to hg19. Dimensionality reduction was achieved using Latent Semantic Analysis (LSA) applied to the filtered peak-barcode matrices. The data was divided into 8 clusters through k-means clustering. Visualization of these clusters was performed using t-distributed stochastic neighbor embedding (t-SNE). For gene annotations, the GENCODE Gene track, version 28lift37, was utilized. For further downstream analysis and visualization of scA-TAC-Seq data, the Loupe Browser, version 6.5.0, was employed. Signac and Seurat R packages were used to complement the analysis. Latent Semantic Indexing (LSI) dimensionality reduction was applied for the Uniform Manifold Approximation and Projection (UMAP) algorithm. Clusters were defined using the Spectral Latent Manifold (SLM) algorithm. Trajectory Inference: To uncover the global structure of the identified clusters and convert this structure into smooth lineages represented by pseudotime, the Slingshot algorithm (Trajectory Inference for Single-Cell Data) was employed.

Gene Set Enrichment Analysis

Differentially expressed genes by RNA-seq were analyzed with Gene Set Enrichment Analysis (GSEA) utilizing hallmark gene sets and single-sample GSEA(ssGSEA) utilizing custom set of 59 genes that make up the ARv7 gene set from Sharp et al.

EnrichR

Enrichment analysis of gene lists was run with a Enrichr, open-source web-based tool. Enrichr is available online at: http://amp.pharm.mssm.edu/Enrichr.

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analyses were carried out using Prism 10 or base R. ANOVA was used to analyze the difference between the means of more than two groups. Unpaired t test was used to compare the mean of two independent groups. Wilcoxon signed-rank test was used for paired comparisons. Wilcoxon rank-sum test was used for unpaired comparisons. Two-sided tests were used for all statistical comparisons. Pearson’s correlation was used to measure the similarity or correlation between two data objects. Statistical significance was as follows: p < 0.05 (*), 0.01 (**), 0.001 (***), 0.0001 (***). Error bars represent the 95th percentile confidence interval.

Highlights.

• AR activity in PC cells treated with AR antagonist (ENZ) is driven by AR splice variant ARv7

• ARv7 activity in ENZ-resistant cells is independent of the full-length AR

• ARv7 function requires adaptations associated with increased chromatin accessibility

• NFI transcription factors are associated with chromatin opening and support ARv7 function